How we use AI

What AI does here

• Monitors sources and groups related news into stories.
• Produces first drafts from the sources we select, following our style guide.
• Suggests headlines, summaries and glossary links.
• Helps check our spelling and grammar.

What a person always does

• Chooses which stories we cover and the angle we take.
• Checks every fact against its source, and removes anything that cannot be verified.
• Adds what AI cannot: hands-on tests, screenshots, interviews, our own view.
• Approves publication. Our publishing system will not publish a draft without an editor's approval.

What AI never does here

• Publish anything on its own.
• Invent quotes, people, test results or sources.
• Appear as a fake author. Bylines are real people; AI is credited as a tool.
• Produce reviews of products we have not tested.

How we label it

Every article ends with a "How this was made" note, for example: "Drafted with AI from the sources listed above, then fact-checked and edited by Davide Serra." Pieces written without AI say so too. Article illustrations are drawn by AI in our house style. They are captioned "AI-generated illustration" and marked in the page's machine-readable data with the IPTC digital source type for AI-made media. We never use AI to make realistic images of real people or events.

Tools we use

We use Claude by Anthropic. A smaller Claude model helps sort which stories are worth explaining, and a larger one drafts the articles and draws the illustrations. We update this list when it changes.

Why we disclose

You should know how what you read was made. Our process follows Google's guidance that AI-assisted content must be reviewed for accuracy, and the EU AI Act, under which AI-generated text on public-interest topics is either labelled or published under human editorial review and responsibility. We do both.

Questions or concerns: info@ai-portal.si.